highmem.c

  1// SPDX-License-Identifier: GPL-2.0
  2/*
  3 * High memory handling common code and variables.
  4 *
  5 * (C) 1999 Andrea Arcangeli, SuSE GmbH, andrea@suse.de
  6 *          Gerhard Wichert, Siemens AG, Gerhard.Wichert@pdb.siemens.de
  7 *
  8 *
  9 * Redesigned the x86 32-bit VM architecture to deal with
 10 * 64-bit physical space. With current x86 CPUs this
 11 * means up to 64 Gigabytes physical RAM.
 12 *
 13 * Rewrote high memory support to move the page cache into
 14 * high memory. Implemented permanent (schedulable) kmaps
 15 * based on Linus' idea.
 16 *
 17 * Copyright (C) 1999 Ingo Molnar <mingo@redhat.com>
 18 */
 19
 20#include <linux/mm.h>
 21#include <linux/export.h>
 22#include <linux/swap.h>
 23#include <linux/bio.h>
 24#include <linux/pagemap.h>
 25#include <linux/mempool.h>
 26#include <linux/init.h>
 27#include <linux/hash.h>
 28#include <linux/highmem.h>
 29#include <linux/kgdb.h>
 30#include <asm/tlbflush.h>
 31#include <linux/vmalloc.h>
 32
 33#ifdef CONFIG_KMAP_LOCAL
 34static inline int kmap_local_calc_idx(int idx)
 35{
 36	return idx + KM_MAX_IDX * smp_processor_id();
 37}
 38
 39#ifndef arch_kmap_local_map_idx
 40#define arch_kmap_local_map_idx(idx, pfn)	kmap_local_calc_idx(idx)
 41#endif
 42#endif /* CONFIG_KMAP_LOCAL */
 43
 44/*
 45 * Virtual_count is not a pure "count".
 46 *  0 means that it is not mapped, and has not been mapped
 47 *    since a TLB flush - it is usable.
 48 *  1 means that there are no users, but it has been mapped
 49 *    since the last TLB flush - so we can't use it.
 50 *  n means that there are (n-1) current users of it.
 51 */
 52#ifdef CONFIG_HIGHMEM
 53
 54/*
 55 * Architecture with aliasing data cache may define the following family of
 56 * helper functions in its asm/highmem.h to control cache color of virtual
 57 * addresses where physical memory pages are mapped by kmap.
 58 */
 59#ifndef get_pkmap_color
 60
 61/*
 62 * Determine color of virtual address where the page should be mapped.
 63 */
 64static inline unsigned int get_pkmap_color(const struct page *page)
 65{
 66	return 0;
 67}
 68#define get_pkmap_color get_pkmap_color
 69
 70/*
 71 * Get next index for mapping inside PKMAP region for page with given color.
 72 */
 73static inline unsigned int get_next_pkmap_nr(unsigned int color)
 74{
 75	static unsigned int last_pkmap_nr;
 76
 77	last_pkmap_nr = (last_pkmap_nr + 1) & LAST_PKMAP_MASK;
 78	return last_pkmap_nr;
 79}
 80
 81/*
 82 * Determine if page index inside PKMAP region (pkmap_nr) of given color
 83 * has wrapped around PKMAP region end. When this happens an attempt to
 84 * flush all unused PKMAP slots is made.
 85 */
 86static inline int no_more_pkmaps(unsigned int pkmap_nr, unsigned int color)
 87{
 88	return pkmap_nr == 0;
 89}
 90
 91/*
 92 * Get the number of PKMAP entries of the given color. If no free slot is
 93 * found after checking that many entries, kmap will sleep waiting for
 94 * someone to call kunmap and free PKMAP slot.
 95 */
 96static inline int get_pkmap_entries_count(unsigned int color)
 97{
 98	return LAST_PKMAP;
 99}
100
101/*
102 * Get head of a wait queue for PKMAP entries of the given color.
103 * Wait queues for different mapping colors should be independent to avoid
104 * unnecessary wakeups caused by freeing of slots of other colors.
105 */
106static inline wait_queue_head_t *get_pkmap_wait_queue_head(unsigned int color)
107{
108	static DECLARE_WAIT_QUEUE_HEAD(pkmap_map_wait);
109
110	return &pkmap_map_wait;
111}
112#endif
113
114unsigned long __nr_free_highpages(void)
115{
116	unsigned long pages = 0;
117	struct zone *zone;
118
119	for_each_populated_zone(zone) {
120		if (is_highmem(zone))
121			pages += zone_page_state(zone, NR_FREE_PAGES);
122	}
123
124	return pages;
125}
126
127unsigned long __totalhigh_pages(void)
128{
129	unsigned long pages = 0;
130	struct zone *zone;
131
132	for_each_populated_zone(zone) {
133		if (is_highmem(zone))
134			pages += zone_managed_pages(zone);
135	}
136
137	return pages;
138}
139EXPORT_SYMBOL(__totalhigh_pages);
140
141static int pkmap_count[LAST_PKMAP];
142static  __cacheline_aligned_in_smp DEFINE_SPINLOCK(kmap_lock);
143
144pte_t *pkmap_page_table;
145
146/*
147 * Most architectures have no use for kmap_high_get(), so let's abstract
148 * the disabling of IRQ out of the locking in that case to save on a
149 * potential useless overhead.
150 */
151#ifdef ARCH_NEEDS_KMAP_HIGH_GET
152#define lock_kmap()             spin_lock_irq(&kmap_lock)
153#define unlock_kmap()           spin_unlock_irq(&kmap_lock)
154#define lock_kmap_any(flags)    spin_lock_irqsave(&kmap_lock, flags)
155#define unlock_kmap_any(flags)  spin_unlock_irqrestore(&kmap_lock, flags)
156#else
157#define lock_kmap()             spin_lock(&kmap_lock)
158#define unlock_kmap()           spin_unlock(&kmap_lock)
159#define lock_kmap_any(flags)    \
160		do { spin_lock(&kmap_lock); (void)(flags); } while (0)
161#define unlock_kmap_any(flags)  \
162		do { spin_unlock(&kmap_lock); (void)(flags); } while (0)
163#endif
164
165struct page *__kmap_to_page(void *vaddr)
166{
167	unsigned long base = (unsigned long) vaddr & PAGE_MASK;
168	struct kmap_ctrl *kctrl = &current->kmap_ctrl;
169	unsigned long addr = (unsigned long)vaddr;
170	int i;
171
172	/* kmap() mappings */
173	if (WARN_ON_ONCE(addr >= PKMAP_ADDR(0) &&
174			 addr < PKMAP_ADDR(LAST_PKMAP)))
175		return pte_page(ptep_get(&pkmap_page_table[PKMAP_NR(addr)]));
176
177	/* kmap_local_page() mappings */
178	if (WARN_ON_ONCE(base >= __fix_to_virt(FIX_KMAP_END) &&
179			 base < __fix_to_virt(FIX_KMAP_BEGIN))) {
180		for (i = 0; i < kctrl->idx; i++) {
181			unsigned long base_addr;
182			int idx;
183			pte_t pteval = kctrl->pteval[i];
184
185			idx = arch_kmap_local_map_idx(i, pte_pfn(pteval));
186			base_addr = __fix_to_virt(FIX_KMAP_BEGIN + idx);
187
188			if (base_addr == base)
189				return pte_page(pteval);
190		}
191	}
192
193	return virt_to_page(vaddr);
194}
195EXPORT_SYMBOL(__kmap_to_page);
196
197static void flush_all_zero_pkmaps(void)
198{
199	int i;
200	int need_flush = 0;
201
202	flush_cache_kmaps();
203
204	for (i = 0; i < LAST_PKMAP; i++) {
205		struct page *page;
206		pte_t ptent;
207
208		/*
209		 * zero means we don't have anything to do,
210		 * >1 means that it is still in use. Only
211		 * a count of 1 means that it is free but
212		 * needs to be unmapped
213		 */
214		if (pkmap_count[i] != 1)
215			continue;
216		pkmap_count[i] = 0;
217
218		/* sanity check */
219		ptent = ptep_get(&pkmap_page_table[i]);
220		BUG_ON(pte_none(ptent));
221
222		/*
223		 * Don't need an atomic fetch-and-clear op here;
224		 * no-one has the page mapped, and cannot get at
225		 * its virtual address (and hence PTE) without first
226		 * getting the kmap_lock (which is held here).
227		 * So no dangers, even with speculative execution.
228		 */
229		page = pte_page(ptent);
230		pte_clear(&init_mm, PKMAP_ADDR(i), &pkmap_page_table[i]);
231
232		set_page_address(page, NULL);
233		need_flush = 1;
234	}
235	if (need_flush)
236		flush_tlb_kernel_range(PKMAP_ADDR(0), PKMAP_ADDR(LAST_PKMAP));
237}
238
239void __kmap_flush_unused(void)
240{
241	lock_kmap();
242	flush_all_zero_pkmaps();
243	unlock_kmap();
244}
245
246static inline unsigned long map_new_virtual(struct page *page)
247{
248	unsigned long vaddr;
249	int count;
250	unsigned int last_pkmap_nr;
251	unsigned int color = get_pkmap_color(page);
252
253start:
254	count = get_pkmap_entries_count(color);
255	/* Find an empty entry */
256	for (;;) {
257		last_pkmap_nr = get_next_pkmap_nr(color);
258		if (no_more_pkmaps(last_pkmap_nr, color)) {
259			flush_all_zero_pkmaps();
260			count = get_pkmap_entries_count(color);
261		}
262		if (!pkmap_count[last_pkmap_nr])
263			break;	/* Found a usable entry */
264		if (--count)
265			continue;
266
267		/*
268		 * Sleep for somebody else to unmap their entries
269		 */
270		{
271			DECLARE_WAITQUEUE(wait, current);
272			wait_queue_head_t *pkmap_map_wait =
273				get_pkmap_wait_queue_head(color);
274
275			__set_current_state(TASK_UNINTERRUPTIBLE);
276			add_wait_queue(pkmap_map_wait, &wait);
277			unlock_kmap();
278			schedule();
279			remove_wait_queue(pkmap_map_wait, &wait);
280			lock_kmap();
281
282			/* Somebody else might have mapped it while we slept */
283			if (page_address(page))
284				return (unsigned long)page_address(page);
285
286			/* Re-start */
287			goto start;
288		}
289	}
290	vaddr = PKMAP_ADDR(last_pkmap_nr);
291	set_pte_at(&init_mm, vaddr,
292		   &(pkmap_page_table[last_pkmap_nr]), mk_pte(page, kmap_prot));
293
294	pkmap_count[last_pkmap_nr] = 1;
295	set_page_address(page, (void *)vaddr);
296
297	return vaddr;
298}
299
300/**
301 * kmap_high - map a highmem page into memory
302 * @page: &struct page to map
303 *
304 * Returns the page's virtual memory address.
305 *
306 * We cannot call this from interrupts, as it may block.
307 */
308void *kmap_high(struct page *page)
309{
310	unsigned long vaddr;
311
312	/*
313	 * For highmem pages, we can't trust "virtual" until
314	 * after we have the lock.
315	 */
316	lock_kmap();
317	vaddr = (unsigned long)page_address(page);
318	if (!vaddr)
319		vaddr = map_new_virtual(page);
320	pkmap_count[PKMAP_NR(vaddr)]++;
321	BUG_ON(pkmap_count[PKMAP_NR(vaddr)] < 2);
322	unlock_kmap();
323	return (void *) vaddr;
324}
325EXPORT_SYMBOL(kmap_high);
326
327#ifdef ARCH_NEEDS_KMAP_HIGH_GET
328/**
329 * kmap_high_get - pin a highmem page into memory
330 * @page: &struct page to pin
331 *
332 * Returns the page's current virtual memory address, or NULL if no mapping
333 * exists.  If and only if a non null address is returned then a
334 * matching call to kunmap_high() is necessary.
335 *
336 * This can be called from any context.
337 */
338void *kmap_high_get(const struct page *page)
339{
340	unsigned long vaddr, flags;
341
342	lock_kmap_any(flags);
343	vaddr = (unsigned long)page_address(page);
344	if (vaddr) {
345		BUG_ON(pkmap_count[PKMAP_NR(vaddr)] < 1);
346		pkmap_count[PKMAP_NR(vaddr)]++;
347	}
348	unlock_kmap_any(flags);
349	return (void *) vaddr;
350}
351#endif
352
353/**
354 * kunmap_high - unmap a highmem page into memory
355 * @page: &struct page to unmap
356 *
357 * If ARCH_NEEDS_KMAP_HIGH_GET is not defined then this may be called
358 * only from user context.
359 */
360void kunmap_high(const struct page *page)
361{
362	unsigned long vaddr;
363	unsigned long nr;
364	unsigned long flags;
365	int need_wakeup;
366	unsigned int color = get_pkmap_color(page);
367	wait_queue_head_t *pkmap_map_wait;
368
369	lock_kmap_any(flags);
370	vaddr = (unsigned long)page_address(page);
371	BUG_ON(!vaddr);
372	nr = PKMAP_NR(vaddr);
373
374	/*
375	 * A count must never go down to zero
376	 * without a TLB flush!
377	 */
378	need_wakeup = 0;
379	switch (--pkmap_count[nr]) {
380	case 0:
381		BUG();
382	case 1:
383		/*
384		 * Avoid an unnecessary wake_up() function call.
385		 * The common case is pkmap_count[] == 1, but
386		 * no waiters.
387		 * The tasks queued in the wait-queue are guarded
388		 * by both the lock in the wait-queue-head and by
389		 * the kmap_lock.  As the kmap_lock is held here,
390		 * no need for the wait-queue-head's lock.  Simply
391		 * test if the queue is empty.
392		 */
393		pkmap_map_wait = get_pkmap_wait_queue_head(color);
394		need_wakeup = waitqueue_active(pkmap_map_wait);
395	}
396	unlock_kmap_any(flags);
397
398	/* do wake-up, if needed, race-free outside of the spin lock */
399	if (need_wakeup)
400		wake_up(pkmap_map_wait);
401}
402EXPORT_SYMBOL(kunmap_high);
403
404void zero_user_segments(struct page *page, unsigned start1, unsigned end1,
405		unsigned start2, unsigned end2)
406{
407	unsigned int i;
408
409	BUG_ON(end1 > page_size(page) || end2 > page_size(page));
410
411	if (start1 >= end1)
412		start1 = end1 = 0;
413	if (start2 >= end2)
414		start2 = end2 = 0;
415
416	for (i = 0; i < compound_nr(page); i++) {
417		void *kaddr = NULL;
418
419		if (start1 >= PAGE_SIZE) {
420			start1 -= PAGE_SIZE;
421			end1 -= PAGE_SIZE;
422		} else {
423			unsigned this_end = min_t(unsigned, end1, PAGE_SIZE);
424
425			if (end1 > start1) {
426				kaddr = kmap_local_page(page + i);
427				memset(kaddr + start1, 0, this_end - start1);
428			}
429			end1 -= this_end;
430			start1 = 0;
431		}
432
433		if (start2 >= PAGE_SIZE) {
434			start2 -= PAGE_SIZE;
435			end2 -= PAGE_SIZE;
436		} else {
437			unsigned this_end = min_t(unsigned, end2, PAGE_SIZE);
438
439			if (end2 > start2) {
440				if (!kaddr)
441					kaddr = kmap_local_page(page + i);
442				memset(kaddr + start2, 0, this_end - start2);
443			}
444			end2 -= this_end;
445			start2 = 0;
446		}
447
448		if (kaddr) {
449			kunmap_local(kaddr);
450			flush_dcache_page(page + i);
451		}
452
453		if (!end1 && !end2)
454			break;
455	}
456
457	BUG_ON((start1 | start2 | end1 | end2) != 0);
458}
459EXPORT_SYMBOL(zero_user_segments);
460#endif /* CONFIG_HIGHMEM */
461
462#ifdef CONFIG_KMAP_LOCAL
463
464#include <asm/kmap_size.h>
465
466/*
467 * With DEBUG_KMAP_LOCAL the stack depth is doubled and every second
468 * slot is unused which acts as a guard page
469 */
470#ifdef CONFIG_DEBUG_KMAP_LOCAL
471# define KM_INCR	2
472#else
473# define KM_INCR	1
474#endif
475
476static inline int kmap_local_idx_push(void)
477{
478	WARN_ON_ONCE(in_hardirq() && !irqs_disabled());
479	current->kmap_ctrl.idx += KM_INCR;
480	BUG_ON(current->kmap_ctrl.idx >= KM_MAX_IDX);
481	return current->kmap_ctrl.idx - 1;
482}
483
484static inline int kmap_local_idx(void)
485{
486	return current->kmap_ctrl.idx - 1;
487}
488
489static inline void kmap_local_idx_pop(void)
490{
491	current->kmap_ctrl.idx -= KM_INCR;
492	BUG_ON(current->kmap_ctrl.idx < 0);
493}
494
495#ifndef arch_kmap_local_post_map
496# define arch_kmap_local_post_map(vaddr, pteval)	do { } while (0)
497#endif
498
499#ifndef arch_kmap_local_pre_unmap
500# define arch_kmap_local_pre_unmap(vaddr)		do { } while (0)
501#endif
502
503#ifndef arch_kmap_local_post_unmap
504# define arch_kmap_local_post_unmap(vaddr)		do { } while (0)
505#endif
506
507#ifndef arch_kmap_local_unmap_idx
508#define arch_kmap_local_unmap_idx(idx, vaddr)	kmap_local_calc_idx(idx)
509#endif
510
511#ifndef arch_kmap_local_high_get
512static inline void *arch_kmap_local_high_get(const struct page *page)
513{
514	return NULL;
515}
516#endif
517
518#ifndef arch_kmap_local_set_pte
519#define arch_kmap_local_set_pte(mm, vaddr, ptep, ptev)	\
520	set_pte_at(mm, vaddr, ptep, ptev)
521#endif
522
523/* Unmap a local mapping which was obtained by kmap_high_get() */
524static inline bool kmap_high_unmap_local(unsigned long vaddr)
525{
526#ifdef ARCH_NEEDS_KMAP_HIGH_GET
527	if (vaddr >= PKMAP_ADDR(0) && vaddr < PKMAP_ADDR(LAST_PKMAP)) {
528		kunmap_high(pte_page(ptep_get(&pkmap_page_table[PKMAP_NR(vaddr)])));
529		return true;
530	}
531#endif
532	return false;
533}
534
535static pte_t *__kmap_pte;
536
537static pte_t *kmap_get_pte(unsigned long vaddr, int idx)
538{
539	if (IS_ENABLED(CONFIG_KMAP_LOCAL_NON_LINEAR_PTE_ARRAY))
540		/*
541		 * Set by the arch if __kmap_pte[-idx] does not produce
542		 * the correct entry.
543		 */
544		return virt_to_kpte(vaddr);
545	if (!__kmap_pte)
546		__kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
547	return &__kmap_pte[-idx];
548}
549
550void *__kmap_local_pfn_prot(unsigned long pfn, pgprot_t prot)
551{
552	pte_t pteval, *kmap_pte;
553	unsigned long vaddr;
554	int idx;
555
556	/*
557	 * Disable migration so resulting virtual address is stable
558	 * across preemption.
559	 */
560	migrate_disable();
561	preempt_disable();
562	idx = arch_kmap_local_map_idx(kmap_local_idx_push(), pfn);
563	vaddr = __fix_to_virt(FIX_KMAP_BEGIN + idx);
564	kmap_pte = kmap_get_pte(vaddr, idx);
565	BUG_ON(!pte_none(ptep_get(kmap_pte)));
566	pteval = pfn_pte(pfn, prot);
567	arch_kmap_local_set_pte(&init_mm, vaddr, kmap_pte, pteval);
568	arch_kmap_local_post_map(vaddr, pteval);
569	current->kmap_ctrl.pteval[kmap_local_idx()] = pteval;
570	preempt_enable();
571
572	return (void *)vaddr;
573}
574EXPORT_SYMBOL_GPL(__kmap_local_pfn_prot);
575
576void *__kmap_local_page_prot(const struct page *page, pgprot_t prot)
577{
578	void *kmap;
579
580	/*
581	 * To broaden the usage of the actual kmap_local() machinery always map
582	 * pages when debugging is enabled and the architecture has no problems
583	 * with alias mappings.
584	 */
585	if (!IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) && !PageHighMem(page))
586		return page_address(page);
587
588	/* Try kmap_high_get() if architecture has it enabled */
589	kmap = arch_kmap_local_high_get(page);
590	if (kmap)
591		return kmap;
592
593	return __kmap_local_pfn_prot(page_to_pfn(page), prot);
594}
595EXPORT_SYMBOL(__kmap_local_page_prot);
596
597void kunmap_local_indexed(const void *vaddr)
598{
599	unsigned long addr = (unsigned long) vaddr & PAGE_MASK;
600	pte_t *kmap_pte;
601	int idx;
602
603	if (addr < __fix_to_virt(FIX_KMAP_END) ||
604	    addr > __fix_to_virt(FIX_KMAP_BEGIN)) {
605		if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP)) {
606			/* This _should_ never happen! See above. */
607			WARN_ON_ONCE(1);
608			return;
609		}
610		/*
611		 * Handle mappings which were obtained by kmap_high_get()
612		 * first as the virtual address of such mappings is below
613		 * PAGE_OFFSET. Warn for all other addresses which are in
614		 * the user space part of the virtual address space.
615		 */
616		if (!kmap_high_unmap_local(addr))
617			WARN_ON_ONCE(addr < PAGE_OFFSET);
618		return;
619	}
620
621	preempt_disable();
622	idx = arch_kmap_local_unmap_idx(kmap_local_idx(), addr);
623	WARN_ON_ONCE(addr != __fix_to_virt(FIX_KMAP_BEGIN + idx));
624
625	kmap_pte = kmap_get_pte(addr, idx);
626	arch_kmap_local_pre_unmap(addr);
627	pte_clear(&init_mm, addr, kmap_pte);
628	arch_kmap_local_post_unmap(addr);
629	current->kmap_ctrl.pteval[kmap_local_idx()] = __pte(0);
630	kmap_local_idx_pop();
631	preempt_enable();
632	migrate_enable();
633}
634EXPORT_SYMBOL(kunmap_local_indexed);
635
636/*
637 * Invoked before switch_to(). This is safe even when during or after
638 * clearing the maps an interrupt which needs a kmap_local happens because
639 * the task::kmap_ctrl.idx is not modified by the unmapping code so a
640 * nested kmap_local will use the next unused index and restore the index
641 * on unmap. The already cleared kmaps of the outgoing task are irrelevant
642 * because the interrupt context does not know about them. The same applies
643 * when scheduling back in for an interrupt which happens before the
644 * restore is complete.
645 */
646void __kmap_local_sched_out(void)
647{
648	struct task_struct *tsk = current;
649	pte_t *kmap_pte;
650	int i;
651
652	/* Clear kmaps */
653	for (i = 0; i < tsk->kmap_ctrl.idx; i++) {
654		pte_t pteval = tsk->kmap_ctrl.pteval[i];
655		unsigned long addr;
656		int idx;
657
658		/* With debug all even slots are unmapped and act as guard */
659		if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL) && !(i & 0x01)) {
660			WARN_ON_ONCE(pte_val(pteval) != 0);
661			continue;
662		}
663		if (WARN_ON_ONCE(pte_none(pteval)))
664			continue;
665
666		/*
667		 * This is a horrible hack for XTENSA to calculate the
668		 * coloured PTE index. Uses the PFN encoded into the pteval
669		 * and the map index calculation because the actual mapped
670		 * virtual address is not stored in task::kmap_ctrl.
671		 * For any sane architecture this is optimized out.
672		 */
673		idx = arch_kmap_local_map_idx(i, pte_pfn(pteval));
674
675		addr = __fix_to_virt(FIX_KMAP_BEGIN + idx);
676		kmap_pte = kmap_get_pte(addr, idx);
677		arch_kmap_local_pre_unmap(addr);
678		pte_clear(&init_mm, addr, kmap_pte);
679		arch_kmap_local_post_unmap(addr);
680	}
681}
682
683void __kmap_local_sched_in(void)
684{
685	struct task_struct *tsk = current;
686	pte_t *kmap_pte;
687	int i;
688
689	/* Restore kmaps */
690	for (i = 0; i < tsk->kmap_ctrl.idx; i++) {
691		pte_t pteval = tsk->kmap_ctrl.pteval[i];
692		unsigned long addr;
693		int idx;
694
695		/* With debug all even slots are unmapped and act as guard */
696		if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL) && !(i & 0x01)) {
697			WARN_ON_ONCE(pte_val(pteval) != 0);
698			continue;
699		}
700		if (WARN_ON_ONCE(pte_none(pteval)))
701			continue;
702
703		/* See comment in __kmap_local_sched_out() */
704		idx = arch_kmap_local_map_idx(i, pte_pfn(pteval));
705		addr = __fix_to_virt(FIX_KMAP_BEGIN + idx);
706		kmap_pte = kmap_get_pte(addr, idx);
707		set_pte_at(&init_mm, addr, kmap_pte, pteval);
708		arch_kmap_local_post_map(addr, pteval);
709	}
710}
711
712void kmap_local_fork(struct task_struct *tsk)
713{
714	if (WARN_ON_ONCE(tsk->kmap_ctrl.idx))
715		memset(&tsk->kmap_ctrl, 0, sizeof(tsk->kmap_ctrl));
716}
717
718#endif
719
720#if defined(HASHED_PAGE_VIRTUAL)
721
722#define PA_HASH_ORDER	7
723
724/*
725 * Describes one page->virtual association
726 */
727struct page_address_map {
728	struct page *page;
729	void *virtual;
730	struct list_head list;
731};
732
733static struct page_address_map page_address_maps[LAST_PKMAP];
734
735/*
736 * Hash table bucket
737 */
738static struct page_address_slot {
739	struct list_head lh;			/* List of page_address_maps */
740	spinlock_t lock;			/* Protect this bucket's list */
741} ____cacheline_aligned_in_smp page_address_htable[1<<PA_HASH_ORDER];
742
743static struct page_address_slot *page_slot(const struct page *page)
744{
745	return &page_address_htable[hash_ptr(page, PA_HASH_ORDER)];
746}
747
748/**
749 * page_address - get the mapped virtual address of a page
750 * @page: &struct page to get the virtual address of
751 *
752 * Returns the page's virtual address.
753 */
754void *page_address(const struct page *page)
755{
756	unsigned long flags;
757	void *ret;
758	struct page_address_slot *pas;
759
760	if (!PageHighMem(page))
761		return lowmem_page_address(page);
762
763	pas = page_slot(page);
764	ret = NULL;
765	spin_lock_irqsave(&pas->lock, flags);
766	if (!list_empty(&pas->lh)) {
767		struct page_address_map *pam;
768
769		list_for_each_entry(pam, &pas->lh, list) {
770			if (pam->page == page) {
771				ret = pam->virtual;
772				break;
773			}
774		}
775	}
776
777	spin_unlock_irqrestore(&pas->lock, flags);
778	return ret;
779}
780EXPORT_SYMBOL(page_address);
781
782/**
783 * set_page_address - set a page's virtual address
784 * @page: &struct page to set
785 * @virtual: virtual address to use
786 */
787void set_page_address(struct page *page, void *virtual)
788{
789	unsigned long flags;
790	struct page_address_slot *pas;
791	struct page_address_map *pam;
792
793	BUG_ON(!PageHighMem(page));
794
795	pas = page_slot(page);
796	if (virtual) {		/* Add */
797		pam = &page_address_maps[PKMAP_NR((unsigned long)virtual)];
798		pam->page = page;
799		pam->virtual = virtual;
800
801		spin_lock_irqsave(&pas->lock, flags);
802		list_add_tail(&pam->list, &pas->lh);
803		spin_unlock_irqrestore(&pas->lock, flags);
804	} else {		/* Remove */
805		spin_lock_irqsave(&pas->lock, flags);
806		list_for_each_entry(pam, &pas->lh, list) {
807			if (pam->page == page) {
808				list_del(&pam->list);
809				break;
810			}
811		}
812		spin_unlock_irqrestore(&pas->lock, flags);
813	}
814}
815
816void __init page_address_init(void)
817{
818	int i;
819
820	for (i = 0; i < ARRAY_SIZE(page_address_htable); i++) {
821		INIT_LIST_HEAD(&page_address_htable[i].lh);
822		spin_lock_init(&page_address_htable[i].lock);
823	}
824}
825
826#endif	/* defined(HASHED_PAGE_VIRTUAL) */